Editorial verification

Every entry on this site goes through an editorial-verification pass before publication, and is re-audited on cadence afterward. The verification state of each entry is published openly on the page itself, at the bottom, in the block titled Editorial verification. That block shows a trust score, the date of the most recent audit, and the count of factual claims that were independently verified.

Synthesis from primary sources, not generation

The prose on this site is synthesised from publicly available source material — declassified documents, court filings, congressional and parliamentary reports, recognised journalism of record, peer-reviewed academic publications. It is not freelanced from a model's internal knowledge, and it is not output that has been published without verification. Synthesis means reading sources and writing summary prose from them. The model functions in the role a research assistant or junior staff researcher would: assembling the published record into a readable account.

The distinction matters because a substantial portion of automatically produced web content is generated without source grounding and published without verification. That pattern produces inaccuracy at scale. The pattern used here is the inverse: synthesis is followed by an independent automated verification pass, and only claims that survive that pass are published.

How verification works

After an entry is drafted, an independent verification orchestrator extracts every factual claim from the text — dates, named persons, named operations, locations, statute citations, court-case citations, agency-renaming events, organisational facts, tenure dates, personnel counts, and any specific factual assertion. Each claim is then verified against the public record using multi-source web search.

Each claim is classified into one of five states:

• Verified — at least two authoritative sources independently confirm the claim.
• Partially verified — one authoritative source confirms; the second is weak or absent, or sources are slightly inconsistent on detail.
• Unverifiable — extensive search did not produce authoritative confirmation, or authoritative sources contradict the claim.
• Contested — authoritative sources actively disagree.
• Closed-source — the claim is verifiable in principle (court filings, FOIA-only documents, paywalled material) but not via open web search.

Claims that come back unverifiable or contested are corrected, replaced with what the public record actually supports, or pulled from the entry. The correction is documented in an immutable audit history. A re-audit confirms that the corrections landed, and the new trust score is published.

Authoritative sources

For verification, the following count as authoritative:

• Primary government documents — official agency materials, congressional and parliamentary records, statutes, court opinions, the Federal Register, GAO and inspector-general reports, declassified documents.
• Court opinions and case citations.
• Recognised news of record — The New York Times, The Washington Post, The Wall Street Journal, The Guardian, the AP, Reuters, the Financial Times, and equivalent national-paper-of-record outlets in the jurisdictions the entry concerns.
• Peer-reviewed academic publications and recognised reference works.
• Recognised think-tank and civil-liberties reports — the Congressional Research Service, the Government Accountability Office, the Brennan Center, the EFF, the ACLU, RUSI, Brookings, RAND, the Carnegie Endowment, and equivalent institutional research.

Wikipedia is not used as the verification — it is used only to locate primary sources. Blogs without primary-source attribution, aggregator sites, and opinion-only articles do not satisfy the verification standard.

The trust score

The trust score displayed on each entry is the percentage of that entry's factual claims that have been independently verified by at least two authoritative sources. The formula is:

trust_score = verified / (verified + partially_verified + unverifiable + contested)

Closed-source claims are excluded from the denominator. Partial verifications do not count toward the numerator. The formula is deliberately strict: it answers the question "what fraction of this entry's factual claims have been independently confirmed against the public record?" rather than a softer "how confident are we overall."

Why all of this is disclosed

The verification state of an entry — its trust score, the date of the most recent audit, the count of claims that were checked — is the most useful single signal a reader can have about whether to rely on a page. Most content on the web is published without independent verification; the writer is the only verifier, and the reader has no way to know which claims were checked and which were not. Disclosing the verification state moves that question from invisible to visible.

This is a methodological commitment, not a claim of perfection. Trust scores below 100% are normal — closed-source items exist, the public record is incomplete on certain topics, and the verifier itself can miss things. The commitment is that the score is honest and is published regardless of whether it is high or low. When an entry cannot be brought to an acceptable trust score, it is pulled until it can be.

For most websites, this method produces more accuracy than the typical publication workflow of a single author writing and self-checking. The exception is the high-resource newsroom with a fact-checking department; this method does not claim parity with that. It claims parity, or better, with the median web reference site.

Provisional scores

Some entries display their trust score with a small (provisional) label after the percentage. A provisional score means: the entry's last full audit found a set of items to fix, those fixes have been applied to the source content, and the score has been recalculated on the assumption that those fixes now verify cleanly — but the independent re-audit that would confirm them has not yet run. A full re-audit costs significant time and compute; we batch them on a slower cadence than the corrections themselves so that fixes can land in public quickly.

The trust score on a confirmed entry means "this fraction of factual claims has been independently confirmed by ≥2 authoritative sources." The trust score on a provisional entry means "this fraction has been confirmed, plus the assumption that the corrections since the last audit landed accurately." The arithmetic is the same; the epistemic strength is not. Provisional scores convert to confirmed scores when the next re-audit runs.

A re-audit can also surface new findings that the initial audit missed — corrections sometimes introduce their own issues, or the verifier catches things on a second pass that the first pass didn't reach. The provisional label is the honest acknowledgement that this work is still pending.

Audit history

Every audit is preserved. The entire history of an entry — every audit's claim-by-claim verification record, every correction applied as a result, every open question that remained — is archived immutably and editorially. The public surface shows only the current verification summary; the underlying record is retained so that the state of each entry's factual claims at any prior date can be reconstructed.

← Back to Methodology